Visual motion. Many slides adapted from S. Seitz, R. Szeliski, M. Pollefeys

Transcription

1 Visual motion Man slides adapted from S. Seitz, R. Szeliski, M. Pollefes

2 Motion and perceptual organization Sometimes, motion is the onl cue

3 Motion and perceptual organization Sometimes, motion is the onl cue

4 Motion and perceptual organization Even impoverished motion data can evoke a strong percept G. Johansson, Visual Perception of Biological Motion and a Model For Its Analsis", Perception and Pschophsics 14, , 1973.

5 Motion and perceptual organization Even impoverished motion data can evoke a strong percept G. Johansson, Visual Perception of Biological Motion and a Model For Its Analsis", Perception and Pschophsics 14, , 1973.

6 Motion and perceptual organization Even impoverished motion data can evoke a strong percept G. Johansson, Visual Perception of Biological Motion and a Model For Its Analsis", Perception and Pschophsics 14, , 1973.

7 Uses of motion Estimating 3D structure Segmenting objects based on motion cues Learning and tracking dnamical models Recognizing events and activities

8 Motion field The motion field is the projection of the 3D scene motion into the image

9 Motion field and parallax X(t) is a moving 3D point Velocit of scene point: V =dx/dt x(t) = (x(t),(t)) is the projection of X in the image Apparent velocit v in the image: given b components v x = dx/dt and v =d/dt These components are known as the motion field of the image X(t) V x(t) v X(t+dt) x(t+dt)

10 Motion field and parallax To find image velocit v, differentiate x=(x,) ( with respect to t (using quotient rule): X(t) V X(t+dt) x = f X Z v x ZVx V = f 2 Z f V x V z x = Z z X v x(t+dt) = f Y Z v = f V V Z z x(t) Image motion is a function of both the 3D motion (V) and the depth of the 3D point (Z)

11 Motion field and parallax Pure translation: V is constant everwhere v x = v = f V f V V Z V x Z z z x v 1 = ( v 0 Vzx), v ( ) 0 = f Vx, f V Z

12 Motion field and parallax Pure translation: V is constant everwhere 1 v = ( v 0 V x), ) v = ( ) z 0 f Vx, f V Z The length of the motion vectors is inversel proportional to the depth Z V z is nonzero: Ever motion vector points toward (or awa from) the vanishing point of the translation direction

13 Motion field and parallax Pure translation: V is constant everwhere 1 v = ( v 0 V x), ) v = ( ) z 0 f Vx, f V Z The length of the motion vectors is inversel proportional to the depth Z V z is nonzero: Ever motion vector points toward (or awa from) the vanishing point of the translation direction V z is zero: Motion is parallel to the image plane, all the motion vectors are parallel

14 Optical flow Definition: optical flow is the apparent motion of brightness patterns in the image Ideall, optical flow would be the same as the motion field Have to be careful: apparent motion can be caused b lighting changes without an actual motion Think of a uniform rotating sphere under fixed lighting vs. a stationar sphere under moving illumination

15 Estimating optical flow I(x,,t 1) I(x,,t) Given two subsequent frames, estimate the apparent motion field u(x,) and v(x,) between them Ke assumptions Brightness constanc: projection of the same point looks the same in ever frame Small motion: points do not move ver far Spatial coherence: points move like their neighbors

16 The brightness constanc constraint I(x,,t 1) I(x,,t) Brightness Constanc Equation: ), ( 1),, ( ),, ( ), ( t x x v u x I t x I + + = ), ( ),, ( ),, ( ), ( Linearizing the right side using Talor expansion: ), ( ), ( ),, ( 1),, ( x v I x u I t x I t x I x t x I v I u I Hence,

17 The brightness constanc constraint I x u + I v + = 0 How man equations and unknowns per pixel? One equation, two unknowns Intuitivel, what does this constraint mean? I The component of the flow perpendicular p to the gradient (i.e., parallel to the edge) is unknown I ( u, v) + I = t t 0

18 The brightness constanc constraint I x u + I v + = 0 How man equations and unknowns per pixel? One equation, two unknowns Intuitivel, what does this constraint mean? I The component of the flow perpendicular p to the gradient (i.e., parallel to the edge) is unknown I ( u, v) + I = t t 0 If (u, v) satisfies the equation, so does (u+u, v+v ) if I ( ( u', v') = 0 gradient (u,v) (u,v ) (u+u,v+v ) edge

19 The aperture problem Perceived motion

20 The aperture problem Actual motion

21 The barber pole illusion

22 The barber pole illusion

23 The barber pole illusion

24 Solving the aperture problem How to get more equations for a pixel? Spatial coherence constraint: pretend the pixel s neighbors have the same (u,v) If we use a 5x5 window, that gives us 25 equations per pixel B. Lucas and T. Kanade. An iterative image registration technique with an application to stereo vision. In Proceedings of the International Joint Conference on Artificial Intelligence, pp , 1981.

25 Solving the aperture problem Least squares problem: When is this sstem solvable? What if the window contains just a single straight edge? B. Lucas and T. Kanade. An iterative image registration technique with an application to stereo vision. In Proceedings of the International Joint Conference on Artificial Intelligence, pp , 1981.

26 Conditions for solvabilit Bad case: single straight edge

27 Conditions for solvabilit Good case

28 Lucas-Kanade flow Linear least squares problem Solution given b The summations are over all pixels in the window B. Lucas and T. Kanade. An iterative image registration technique with an application to stereo vision. In Proceedings of the International Joint Conference on Artificial Intelligence, pp , 1981.

29 Lucas-Kanade flow Recall the Harris corner detector: M = A T A is the second moment matrix We can figure out whether the sstem is solvable b looking at the eigenvalues of the second moment matrix The eigenvectors and eigenvalues of M relate to edge direction and magnitude The eigenvector associated with the larger eigenvalue points in the direction of fastest intensit change, and the other eigenvector is orthogonal to it

30 Interpreting the eigenvalues Classification of image points using eigenvalues of the second moment matrix: λ 2 Edge λ 2 >> λ 1 Corner λ 1 and λ 2 are large, λ 1 ~ λ 2 λ 1 and λ 2 are small Flat region Edge λ 1 >> λ 2 λ 1

31 Uniform region gradients have small magnitude small λ 1, small λ 2 sstem is ill-conditioned

32 Edge gradients have one dominant direction large λ 1, small λ 2 sstem is ill-conditioned

33 High-texture or corner region gradients have different directions, large magnitudes large λ 1, large λ 2 sstem is well-conditioned

34 Errors in Lucas-Kanade The motion is large (larger than a pixel) Iterative refinement Coarse-to-fine estimation Exhaustive neighborhood search (feature matching) A point does not move like its neighbors Motion segmentation Brightness constanc does not hold Exhaustive neighborhood search with normalized correlation

35 Feature tracking So far, we have onl considered optical flow estimation in a pair of images If we have more than two images, we can compute the optical flow from each frame to the next Given a point in the first image, we can in principle reconstruct its path b simpl following the arrows

36 Tracking challenges Ambiguit of optical flow Need to find good features to track Large motions, changes in appearance, occlusions, disocclusions Need mechanism for deleting, adding new features Drift errors ma accumulate over time Need to know when to terminate a track

37 Tracking over man frames Select features in first frame For each frame: Update positions of tracked features Discrete search or Lucas-Kanade (or a combination of the two) Terminate inconsistent tracks Compute similarit with corresponding feature in the previous frame or in the first frame where it s visible Find more features to track

38 Shi-Tomasi feature tracker Find good features using eigenvalues of secondmoment matrix Ke idea: good features to track are the ones whose motion can be estimated reliabl From frame to frame, track with Lucas-Kanade This amounts to assuming a translation model for frame-to-frame feature movement Check consistenc of tracks b affine registration to the first observed instance of the feature Affine model is more accurate for larger displacements Comparing to the first frame helps to minimize drift J. Shi and C. Tomasi. Good Features to Track. CVPR 1994.

39 Tracking example J. Shi and C. Tomasi. Good Features to Track. CVPR 1994.